[lxc-devel] detecting reboot/halt...

[Daniel Lezcano]

Michael Tokarev wrote:
> I come across a series of patches to implement
> reboot/halt of a container.  Patches were discussed
> before, but I had no time to look at that stuff in
> more detail...


> The problem.  The current detection is based on the
> content of container's /var/run/utmp.  This is goood
> provided the container actually touches that file,
> but this is not true for "single-application"
> containers, i.e. the ones without full init.d
> system.

Mmmh, yes that's something I thought about but I assumed an application 
container will not do 'shutdown' or/and run as root. But maybe I missed 
one kind of application, do you have in mind a particular one ?

> There are only two reliable ways to get this to work:
> it's either catching sys_reboot() (for which kernel
> support/changes is needed), or replacing poweroff or
> halt (or introducing a separate utility) which will
> notify the parent process about the event in some
> way (sending signal, connecting to a socket, writing
> something to a pipe and the like).

The first solution (catch at the kernel level) is a better solution 
covering all the cases. Adding extra utilities is not easy to manage 
because of the OS update and package dependencies.

> I remember something of that sort were proposed
> too, but I can't find the details now.  Are there
> some patches or git branch with that stuff?

Yep, I did a quick hack in sys_reboot sending a SIGPWR to the parent of 
the pid namespace when this one is not the init_pid_ns, but I didn't had 
time to propose/send an acceptable version and argue for it on 
lkml@/containers at . Furthermore, before sending a patch to the kernel, we 
have to be sure this problem can't be solved from userspace first.

Thanks
   -- Daniel